Sizing a NFS cluster
 

Hi all

We need to build a HA NFS cluster (2 VMs) and need some help sizing this please.

The NFS server(s) will be fiber connected to HP Primera SAN. There will be around 5TB of attached storage in total, the largest volume being just under 2TB and around a dozen filesystems between 50 and 200GB and a couple slightly larger.

None of these are high throughput I/O - the application team have estimated 16 MiB/s being the max required, based on some experience with a similar system using Azure NetApp Files (this one won't be in Azure so can't use ANF).

Most of the 50GB shares will only be connected to between 2 and 6 VMs. The larger one and some others may need to be connected to between 50, 60 VMs or so.

Based on several articles / examples I've been able to find, I was thinking that perhaps 2 x 32GB RHEL VMs with 4 cores be sufficient?

Any thoughts / feedback much appreciated

Thanks
Ross

From experience, requirements can change. Not a whole lot you can do about that.
But to answer your question; again from experience, I had HA NFS cluster for 124 nodes with high IO thruput requirement (using IB). The 2 nodes serving NFS were 8 cores and 64GB RAM and worked without a hitch.
Since your requirements are about half the size and not high thruput, I am going to guess yes 4 CPUs and 8GB RAM *should* work.
Good thing is your NFS servers are VMs, so you can easily up those numbers.
Assuming you have experience with NFS, then you should know that tuning NFS is part of getting better performance and not just adding more CPU/memory to the servers.

Can't you just expose volumes on the SAN as NFS mounts then the relevant VMs can connect directly without the need for an intermediate NFS cluster.

Thank you both, much appreciated.

The SAN does not natively support NFS (and even if it did, we have to go with what our hosting partner offers) hence we need to build an NFS service.